Expressive Text-to-Image Generation with Rich Text

Songwei Ge, Taesung Park, Jun-Yan Zhu, and Jia-Bin Huang
International Conference on Computer Vision (ICCV), 2023

📝 Paper: https://arxiv.org/abs/2304.06720
🌐 Website: https://rich-text-to-image.github.io/
💻 Code: https://github.com/SongweiGe/rich-text-to-image
🤗 Demo: https://huggingface.co/spaces/songweig/rich-text-to-image
🖥️ A1111 extension: https://github.com/songweige/sd-webui-rich-text

📄 Abstract: Plain text has become a prevalent interface for text-to-image synthesis. However, its limited customization options hinder users from accurately describing desired outputs. For example, plain text makes it hard to specify continuous quantities, such as the precise RGB color value or importance of each word. Furthermore, creating detailed text prompts for complex scenes is tedious for humans to write and challenging for text encoders to interpret. To address these challenges, we propose using a rich-text editor supporting formats such as font style, size, color, and footnote. We extract each word's attributes from rich text to enable local style control, explicit token reweighting, precise color rendering, and detailed region synthesis. We achieve these capabilities through a region-based diffusion process. We first obtain each word's region based on attention maps of a diffusion process using plain text. For each region, we enforce its text attributes by creating region-specific detailed prompts and applying region-specific guidance, and maintain its fidelity against plain-text generation through region-based injections. We present various examples of image generation from rich text and demonstrate that our method outperforms strong baselines with quantitative evaluations.

Jia-Bin Huang

A research approach enabling expressive text-to-image generation using rich text formatting (font styles, colors, sizes, footnotes, hyperlinks) as input. By extracting token maps from cross-attention and self-attention maps in diffusion models, the system achieves localized image editing tied to specific text tokens. Techniques include spectral clustering for precise spatial masks, weighted noise averaging per token, region-based gradient guidance for RGB color control, and feature map injection to maintain structural consistency with plain text generation. Demonstrated on Stable Diffusion XL with examples like changing hair color, applying artistic styles to specific objects, and referencing external images via hyperlinks.

Expressive Text-to-Image with Rich Text